An Improved Framework for Layering Linguistic Processes in Sign Language Generation: Why There Should Never Be a “Brows” Tier

A survey on the animation of signing avatars: From sign representation to utterance synthesis

“…Addingnoise, ambient motions, autonomous behaviors ensuring breathing motions, or modifying the timing of the motion on the different channels are recurrent methods used to improve the realism of avatar motions. Signal processing techniques based on motion capture data analysis can also be used to improve the realism of avatar motions [58].Examples of avatars using procedural means to add realism:• In the ViSiCAST project[74], noise, damping and ambient motions (small random eye / head / torso motions) have been added to reduce the robotic and unnaturally stiff movement inherent to their dynamics-based model.• A small de-synchronization of the hands during the performance of symmetrical signs has been added to the Irish Sign Language avatar of Smith et al[75].• In the EMBR project of Héloir & Kipp [52], an autonomous behaviour ensuring breathing and some natural movements like blinking as well as blushing effects has been implemented.• To liven the avatar Paula, noise has been added to its movements[76].However, even if various techniques are implemented to overcome the limitations of synthesis techniques based on synthetic models, they can not compete with data-driven synthesis techniques in terms of realism.4.3.3. Data-Driven Sign SynthesisData-driven synthesis techniques use captured motions to animate an avatar.…”